Field of Invention
The present invention relates to technical fields of silicon-based photonic integrated device combined with neuromorphic chip, and more particularly to an optical-readout synaptic device based on SiOxNy and a preparation method thereof.
Description of Related Arts
For a memristor having a sandwich structure of metal/(dielectric layer)/metal, if being exerted by different bias voltages, the resistance value of the device will change nonlinearly. The nonlinear change of the resistance is caused by the formation or disappearance of the electric channel in the dielectric layer under the different bias voltages. However, the connection strength of the nanoscale filamentous electric channel will change with the amplitude and the action time of the bias voltage. Interestingly, this operation characteristic is similar as the working mechanism of the synapses connected with the different nerve cells in the biological neural system. Because of the similarity between the memristor and the synapses of the biological system, the memristor is suitable to serve as the synaptic device for constructing the neuromorphic bionic-brain chip and further for the artificial neural network. It is confirmed by the researches that the bionic synapse based on the memristor is the nearest bionic device to the synapses in the biological neural system until now.
The memristor provides an excellent bionic synaptic device for the artificial neural network. However, the conventional bionic synaptic device based on the memristor reads the synaptic weight of the synaptic device with the electric signal, and the process is called the electrical-readout. The electrical-readout memristive bionic synapse has the small transmission bandwidth and the large interference among the signals. In comparison, reading the synaptic weight with the optical signal (amplitude and phase) as the information medium, which is called the optical-readout, has the incomparable advantages over the electrical-readout, such as the large transmission bandwidth and the excellent parallel signal processing capacity. However, until now, all of the published related documents about the synaptic device based on the memristor effect are based on the electrical-readout bionic synapse.
The Chinese patent publication of CN 104916313A disclosed a neural network synaptic structure based on electrical-readout memristive devices, comprising a first memristive device, a second memristive device, a voltage inverter, an operational amplifier and a feedback resistance.
The Chinese patent publication of CN 105287046A disclosed a full carbon-based electrical-readout neural synaptic bionic device and a production method thereof. The full carbon-based neural synaptic bionic device comprises a substrate, a bottom electrode on the substrate, a resistive functional layer on the bottom electrode and a top electrode on the resistive functional layer, wherein the bottom electrode and the top electrode are made of graphene; and the resistive functional layer is made of graphene oxide.
The Chinese patent publication of CN 105304813A disclosed an electrical-readout neural synaptic bionic device and a preparation method thereof. The neural synaptic bionic device comprises a substrate, a bottom electrode on the substrate, a resistive functional layer on the bottom electrode and a top electrode on the resistive functional layer, wherein the resistance transformation layer comprises an upper oxide layer, a lower oxide layer, and metal nanoparticles between the upper and lower oxide layers.
The Chinese patent publication of CN 104934534A disclosed an electrical-readout biological neural synaptic bionic electron device and a preparation method thereof. The biological neural synaptic bionic electron device from top to bottom successively comprises an upper electrode, a middle insulation layer connected with the upper electrode and a lower electrode connected with the middle insulation layer, wherein: the upper electrode is made of titanium; and the middle insulation layer is made of metallic oxide.
The Chines patent publication of CN 104376362B disclosed an electrical-readout synaptic device for the artificial neural network. The artificial neural network comprises multiple nerve cells; the nerve cells are connected with each other through the synaptic device array; the synaptic device array comprises multiple synaptic devices; and each synaptic device comprises at least two synaptic units which are connected in parallel.
The present invention provides a bionic synaptic device based on the optical-readout.